Modeling the Deployment and Management of Large-Scale Autonomous Vehicle Circulation in Mixed Road Traffic Conditions Considering Virtual Track Theory
1. Introduction
With the advent of autonomous driving and its interaction with connected intelligent transportation systems, the promotion of an “intelligent” road vehicle network with an efficient management and control strategy for autonomous vehicles has become an essential part of autonomous mobility research. Higher requirements and challenges are put forward for future autonomous transportation infrastructures, the autonomous vehicle management and control systems, the vehicle-infrastructure cooperated autonomous driving (VICAD) services, and safety rules. The existing autonomous driving technologies are mainly focused on single vehicle intelligence and tend to ignore the wider network-related problems and challenges related to the overall movement of autonomous vehicles on the network, i.e., the macro-level. There are still problems and issues associated with autonomous driving when dealing with these challenges. Basically, there are two main areas of consideration:
-
The safety and stability of autonomous vehicles as seen at the micro-level, i.e., the processing of the individual vehicle movement in the road network, and
-
The safety and stability of the traffic stream at the macro-level, which includes the movement of the autonomous vehicles in the traffic stream together with other human-driven vehicles.
The research reported in this paper focuses on both these areas, but primarily on the second, as it considers the movement and control of autonomous vehicles within a full network at mixed traffic conditions. It is doing so by presenting a novel methodology for converting the autonomous road vehicle movement to a railway-like system in which the road travel scene of autonomous vehicles is transformed into a railway travel scene. In this way, conditions are created for the application of railway train operation and control ideas onto the autonomous road transport operation. In doing so, the proposed “virtual track” theory application is based on simulating the movement of a batch of autonomous vehicles on the road network to the movement of a train on the rail network, and then applying train management and control strategies and algorithms. This idea can be a game changer since autonomous vehicles do use an extensive array of various sensors, network controllers, cloud-based services and resource databases, etc., that make their movement subject to—or particularly suited—to a “guided” way of operation, similar to that of railway vehicles. Data perception, data transmission, and independent decision-making are the main characteristics of smart rail vehicles that can eliminate security risks at the vehicle end. New technologies and business models, such as the use of artificial intelligence, big data, cloud computing, and the features of the open Internet can better be applied to a concept that resembles rail vehicle movement, i.e., movement following a track, rather than that of a free moving road autonomous vehicle.
The strengthening of the safety and stability of the operation of autonomous vehicles, and the lessening of the risks involved in the operation of autonomous road vehicles by simulating their movement to match that of a rail vehicle is strengthened by the following further considerations. The inner software of a road autonomous vehicle has a congenital safety hazard itself in that, at the beginning of its design, the vehicle’s safety decisions vis a vis the hazards it can face cannot be manifested clearly, and they cannot be compared to those perceived by an independent driving individual at present. These decision rules will be improved with the continuous deepening of vehicle intelligence, networking, and the interaction with external data. All this, together with the tendency of vehicles to become more intelligent and more connected, will make automobiles increasingly complex and more computer-like with software codes that exceed hundreds of millions of lines and a great number of sensors and Electronic Control Units (ECUs). The issues of successfully resolving the safety hazards of the vehicle under a large number of situations (application scenarios) will become increasingly prominent and, at the end, may perhaps be the main concern of autonomous vehicles manufacturers. The emergence of certain specific scenarios will always cause a potentially fatal safety hazard for the operation of autonomous vehicles, and this danger is reduced when equating the movement of road autonomous vehicles with that of trains moving on a rail track.
Existing autonomous driving technologies primarily focus on the individual vehicles and the use of advanced sensors (radar, camera), controllers, actuators, and vehicle-mounted sensing systems and information terminals to convert them to autonomous driving. These technologies will need to be complemented with network-wide applications of intelligent information exchange between the users (people), the vehicles, the roads, and the control centers to further enable autonomous vehicles to move safely within the traffic stream and acquire intelligent environmental perception and other capabilities. The ultimate goal is to make them able to face the many dangerous situations and states that can arise in a free and unchecked by a human driver movement in accordance with the user’s wishes. The method of autonomous traffic control at the macro-level proposed in this paper addresses the above considerations and enables full and effective control of their movement by virtually converting the autonomous road vehicle movement to an equivalent movement of a railway convoy of vehicles (train). So, the research questions to which this paper will attempt to answer, are:
-
Can we transform the autonomous road vehicles’ operation into a railway-like one?
-
Is the “virtual track” theory and its related mathematics a practical tool for converting road autonomous vehicles to a train-like process?
-
How can we optimize and control the movement of the “virtual track” autonomous vehicles?
-
How feasible and practical is such a transformation?
1.1. Literature Review
The focus of this paper is the macro-scale management and control of autonomous vehicles circulating in the traffic stream together with other human-driven vehicles. The basic concept is to replicate (simulate) the movement of autonomous vehicles with that of railway wagon formations, and then apply railway train operation control techniques to guide highway autonomous driving. The literature review therefore involves a wide range of fields which have been grouped in four categories starting with the initial basic vehicle routing problems, then moving on to the more recent work on vehicle formation, traffic control, dynamic traffic allocation simulation, and finally, to the vehicle following theories and vehicle trajectory optimization models. At the end, the relevant literature about the theory and methods of rail transport and train operation control, which are the most relevant issues to the virtual track theory application used in this paper, are reviewed.
1.1.1. Vehicle Routing, Path Planning and Dispatching
1.1.2. Vehicle Formation, Traffic Control and Dynamic Traffic Allocation
1.1.3. Vehicle following Models
1.1.4. Vehicle Trajectory Optimization Models
1.1.5. Optimization of Train Operation Diagram
As this paper adopts the method of transforming an autonomous vehicle highway scene into a corresponding railway scene, known as “highway virtual track”, it is of interest to also review some literature from the theory and methods of rail transport and train operation control. The train operation diagram is a technical document that describes the train movement as it runs on the railway track as well as the time stopping or passing at the stations. It is the basis for organizing train operation and constitutes an essential starting point for railway train operation control. When studying the temporal-spatial trajectory optimization and system-related functional modules, the relevant references are those referring to the optimization of the train operation diagram and the applicability of the results of such optimizations.
2. The Need for “Safety of Life” Security Level of Autonomous Vehicle Operation
The “fail safe” operation required for autonomous vehicle movement can only be secured by implementing rigorous movement control systems aided by rigorous vehicle to infrastructure (V2I) and (V2X) communication and cooperation protocols as well as strict safety standards. As regards the safety standards, the second edition of the American National Standards Institute’s—ANSI/UL 4600 Standard for Safety for the Evaluation of Autonomous Products was published in March 2022; it provides a way to assess the safety case of autonomous vehicles and is one of the few comprehensive standards for public road autonomous vehicle safety that covers both urban and highway use cases. Other notable relevant safety standards are the ISO 26262—Road Vehicles Functional Safety Package, and the ISO 21448 road vehicles safety of the intended functionality standard of 2022 (SOTIF) that refers to hazards caused by functional insufficiencies.
As regards the vehicle to infrastructure (V2I) and (V2X) communication/cooperation, the “virtual track” operation suggested in this paper is a most notable advance; its feasibility and technical characteristics are analyzed in the following sections.
3. Presentation of the Virtual Track Theory
3.1. The Need for Vehicle-Road Collaboration in Autonomous Driving
Vehicle-road collaborative autonomous driving (V2I) is a three-in-one automated driving technology, i.e., one that combines the capabilities of three new generations of telecommunications, i.e., 5G or 6G, the physical Internet, and artificial intelligence (AI) technologies. Through real-time dynamic information and data exchange between the vehicles and the infrastructure (roads and control centers), the collection and analysis of temporal-spatial dynamic traffic information is achieved, as well as its integration in the active safety control of autonomous vehicles and the management of their movement. Implementation of the concept of virtual track further enhances and realizes the effective cooperation between the infrastructure and vehicles, ensures traffic safety, improves traffic efficiency, and forms an intelligent, safe, and efficient road traffic system.
The above considerations apply primarily to the macro-scale, i.e., road traffic situations such as urban road networks or highways, and not to single vehicle considerations.
3.2. The Virtual Track Concept
The core idea of the virtual track theory is to transform the road travel scene of autonomous vehicles into railway travel scenes, and then control the operation of autonomous road vehicles like railway trains. To do this, the movement of autonomous vehicles on a road traffic situation (scenarios), needs to be converted into a railway-like track convoy level. The virtual track method can be used to do this and enable the effective and, above all, safe control of any large-scale autonomous vehicles’ circulation in complex road traffic scenarios.
The virtual track provides data and information to autonomous vehicles through the roadside communication infrastructure, allowing autonomous vehicles to better coordinate with other vehicles and the road environment. In this way the operational safety of autonomous vehicles is improved as they operate on complex physical road networks and, overall, achieve more precise positioning and navigation of autonomous vehicles. Urban roads or intercity highways can be virtually tracked by making one lane to correspond to a virtual track. On the virtual lane-track, multiple autonomous vehicles are queued to form a vehicle formation similar to a train and controlled to follow a planned trajectory to further improve the operation safety of autonomous vehicles in road traffic environments.
3.3. Virtual Track Construction Process
The scheduling or dispatching nodes play the role of the (rail) stations and are usually taken to coincide with the highway service areas that are distributed at unequal intervals on the highway network. These play the role of hubs and traffic flow nodes similar to the train stations in a railway system. Therefore, node processing for the virtual track transformation is performed on the highway based on the service area locations.
3.4. Combining Controlling Sections into a Network
Before this sequence starts to be implemented, the system needs to determine the “occupancy locking” situation of the receiving controlling cells through signal information exchange and V2X and V2I information exchange communication. The autonomous vehicle control system collects and processes the necessary data. If the receiving controlling cell is not occupied by another autonomous vehicle, that is if it is not “locked”, the turning autonomous vehicle can enter this cell and corresponding section, otherwise a delay occurs, and the path needs to be readjusted.
After the interval distribution, the node processing, and the virtual turnout settings are completed, the virtual-track scene of the whole highway is set, and the transformation of the road travel scene to a railway travel scene has been completed. At that stage, the railway train operation control ideas and methods can be used to manage and control the autonomous vehicles movement on the highway. Because, in the virtual track, the lanes of the highway are composed of the controlling cells, that is of multiple units, the timetable plan and the method of running graphs of railway train operation control can be used. Under the movement operation in virtual tracks, the safety of autonomous vehicles no longer relies only on the perception, planning, and decision-making of the individual vehicle, but also relies on the overall organizational scheduling and control of a railway-like control center.
In addition, through the signal control and vehicle queue mechanism under the virtual track conditions, the autonomous vehicles running on the highway can be grouped into formations and use integrated autonomous vehicle control system algorithms and accident safe controls.
3.5. “Virtual Track” Autonomous Vehicle Management and Control
4. Model Construction and Testing Based on Virtual Track Concept
4.1. Symbol Definition
4.2. Construction of the Model
4.2.1. Constraint Conditions
The autonomous vehicle management and control model that is constructed for the optimization of the virtual track-based movement of autonomous vehicle platoons on a road network has to respect the following constraints:
By introducing the above constraints, the proposed model ensures that the application of the virtual track theory is in full alignment with the safety and effectiveness of autonomous vehicle movement and control.
4.2.2. Model Formulation and Objective Function Confirmation
This model is an integer programming model, and optimization software such as the CPLEX solver from IBM ILOG can be used to solve it. This is a high performance solver for Linear Programming (LP), Mixed Integer Programming (MIP) and Quadratic Programming (QP/QCP/MIQP/MIQCP) problems. However, solution difficulties may arise when the number of relevant variables is large.
4.3. Model Verification and Solution in a Simplified Network
5. Conclusions and Further Work
This paper has investigated the case of safe and efficient operation of autonomous road vehicles moving on an urban or interurban road network on the basis of virtual track theory. The proposed methodology transforms the autonomous road vehicles’ operation into a railway-like one by converting the road space into “cells” occupied by one vehicle at the time and then grouping these cells into virtual train-like formations that are then managed and controlled like a train scheduled to go from origin point to a destination passing through intermediate “stations” that are the so-called dispatching nodes that, in this paper, are suggested to coincide with the service stations along a highway. The preceding analysis has demonstrated that such transformation is feasible, and there are practical software tools, supported by the proper mathematics, for effecting this conversion from a road movement scene to a railway one. It also demonstrated how one can optimize the whole process and, in this way, provide an efficient and, above all, safe management and control process for the movement of autonomous vehicles on the road networks of the future. The demonstration of the application of the proposed method has—for this case—been made on a hypothetical small-scale network pending a bigger full-scale demo on a real highway network that is to be performed at a next stage.
The full-scale application of the proposed methodology is therefore the first item of future work that must be stressed. There are, however, other improvements that can be mentioned here. In this paper, we used the minimum operating cost of an autonomous vehicle as the target function to minimize when the model is constructed. In the current phase of building the model, the setting of this target function is relatively simple. In the future practical applications in real world situations, this may not be enough, and more target functions could be necessary to be considered. These may include more complicated factors such as the utilization rate of road capacity or overall delays and congestion costs, or generalized costs, etc. Diverse issues like road maintenance time windows would also have to be considered. In future research in this area, therefore, one would need to comprehensively consider multiple factors for inclusion into the optimization process to further improve and supplement the model.
Also, as regards the constraint conditions that were used, the emphasis was given here to constraints that focused on the safety of the vehicles on the virtual-tracked highway. Other constraints that could be tested include the adaptability of the road controlling cells formations to the running control of railway trains, or to the macro-level organizational management and scheduling control of large-scale autonomous vehicle formations. In this way, the application of the virtual track theory will not only ensure the safety but also the effectiveness of the autonomous vehicles’ movement and their management and control through better and more versatile models and algorithms.
Another possible improvement for the future of virtual track theory could be the application of a three-layer conversion of the road network to “macro–meso–micro” levels instead of the current dual-layer. The virtual track theory applied in this paper can carry out macro and micro two-layer transformation of road network, but there is a lack of a layer of transition between macro and micro transformation of road network, which is not conducive to improving the controlling accuracy of autonomous vehicles. In future research, the virtual track theory will be further developed to achieve the goal of “macro–meso–micro” three-layer transformation of the road network, so as to make the autonomous vehicle control scheme more accurate and of high-precision.
Our overall conclusion is that the case of the virtual track theory is a promising avenue for research, offering a convenient railway travel scene in the place of road one by way of simple transformation steps that also include the possibility of (road) vehicle formation into platoons as a way/tool to form something like a train. Once the road movement of autonomous vehicles in the traffic stream has been transformed into a railway-like movement on a track, the use of railway scheduling and solution tools can be used, as demonstrated in this paper. This approach may offer a new angle for autonomous vehicle movement research in the future which, together with the expected progress in high power cloud-computing, super-computing and super-fast information communication (V2X), can make the train-like management and control of the autonomous vehicles (even those running on a mixed traffic scene) both feasible and effective. This paper has focused on methodological issues rather than on models and algorithms as an introduction of the thinking of autonomous driving from the perspective of railway train control. Its preliminary results are encouraging and convincing enough to secure further development and a first real world and larger scale application. This research will be continued, and the theory of virtual tracks will be further developed to hopefully provide an easy and accurate tool for the management and control of autonomous vehicles in our road networks.